SRL-TR²: A Safe Reinforcement Learning Based TRajectory TRacker Framework

Safety in Traffic Management Systems: A Comprehensive Survey

“…Similarly, Du et al [19] integrated domain rules into an existing backbone RL model to enhance safety at intersections. Wang et al [18] applied RL methods coupled with safety constraints and expert strategies for the trajectory-tracking control problem. Das et al [11] introduced a dual RL agent-based method to achieve an optimal tradeoff between traffic efficiency and driving safety/comfort.…”

“…In the field of autonomous driving, there is a general preference for models with transferability, primarily due to the fact that a significant portion of the existing literature relies on simulated environments rather than real-world scenarios. Wang et al [18] achieved impressive results by successfully transferring their one-shot learning approach across simulation and realistic scenarios, showcasing a low average running time and minimal lateral error during field tests. Building on this, Wang et al [55] investigated the varying influences of driving patterns exhibited by surrounding vehicles in different positions on the evaluation of driving risk, highlighting the strong influence of cross positions followed by diagonal cross positions.…”

“…Many researchers highlight the difficulty in collecting labeled accident data in real-world scenarios [10]. Another common issue is the lack of generalizability to real-world conditions [18], as some proposed models demonstrate satisfactory performance only in simulated environments, with limited evidence of successful deployment in real-world settings. It is essential to recognize and address these challenges in order to advance the field of traffic safety and improve the applicability of the research findings in practical contexts.…”